• Data Science technieken op de dataset toepassen (exploratief)
• Vooraf geen duidelijke businescase en/of probleemstelling
• Vooral toepassen en wellicht verrassen(?)
• Uitwerking met focus op prestatieveld huurdersoordeel 2019

Git Repository: https://github.com/DekkerMartijn/AedesBenchmark

length(Df_Huurderoordeel_jaar$Corporatie)

## [1] 381

prop.table(table(Df_Huurderoordeel_jaar$Letter.huurdersoordeel..Letters..3.klasse..))

## 
##         A         B         C 
## 0.2242647 0.4963235 0.2794118

scatterplotMatrix(Model_Df[,c(2:4)])

plot_ly(Model_Df, x=~X_R, y=~X_NH, 
             z=~X_VH, color=~Y) %>%
  add_markers(size=1.5)

Model_Df5$cluster = factor(kmeans(Model_Df5,3)$cluster)
levels(Model_Df5$cluster) <- c("A","B","C")

##       fitted
## actual  A  B  C
##      A 58  2  1
##      B 59 42 34
##      C  0 52 24

# Fit tree model
set.seed(100481)
Tree <- rpart(Y~., data=Model_Df)
rpart.plot(Tree, box.palette="RdBu", shadow.col="gray", nn=TRUE)

# Confusion Matrix
fit.val <- predict(Tree, type = "class")
table(actual = Model_Df$Y, fitted = fit.val)

##       fitted
## actual   A   B   C
##      A  56   5   0
##      B   0 134   1
##      C   0   8  68

rf <- randomForest(
  Y ~ .,
  data=Model_Df, importance=TRUE
)
rf

## 
## Call:
##  randomForest(formula = Y ~ ., data = Model_Df, importance = TRUE) 
##                Type of random forest: classification
##                      Number of trees: 500
## No. of variables tried at each split: 1
## 
##         OOB estimate of  error rate: 1.1%
## Confusion matrix:
##    A   B  C class.error
## A 59   2  0 0.032786885
## B  0 134  1 0.007407407
## C  0   0 76 0.000000000

indices <- vector(mode = "logical", length = length(Model_Df$Y))
for (i in 1:length(Model_Df$Y)) { if (Model_Df$Y[i] == rf$predicted[i]) 
    { indices[i] <- TRUE } 
      else indices[i] <- FALSE
}
Check_False <- Model_Df[which(!indices),]
Check_False$Predicted <- rf$predicted[which(!indices)]Check_False

##                 Y X_NH X_R X_VH Predicted
## Heuvelrug Wonen A  8.1 7.9  7.1         B
## Stadgenoot      B  7.6 7.1  8.0         C
## Wonen Delden    A  7.6 7.9  7.7         B

rf2 <- randomForest(
  Y ~ .,
  data=Model_Df4, importance=TRUE
)rf2

## 
## Call:
##  randomForest(formula = Y ~ ., data = Model_Df4, importance = TRUE) 
##                Type of random forest: classification
##                      Number of trees: 500
## No. of variables tried at each split: 8
## 
##         OOB estimate of  error rate: 39.84%
## Confusion matrix:
##   A  B  C class.error
## A 0 16  2   1.0000000
## B 0 46 14   0.2333333
## C 0 17 28   0.3777778

varImpPlot(rf2, n.var = 10, type=1)

• Beoordelingskader
• Inlvoed van overige kenmerken
• Andere technieken?
• Suggesties